Merge branch 'master' of https://github.com/COSMOGRAIL/PyCS

demo/demo1/access_results.py

@@ -0,0 +1,18 @@
+
+import pycs
+
+results = pycs.sim.run.collect(directory="./sims_copies_opt_spl")
+
+# The object "results" is an instance of the class pycs.sim.run.runresults. You can directly access the following attributes:
+
+print results.labels # A list of the QSO image names (defines the order of QSO images with which the following results are given)
+print results.tsarray # A 2D array with the measured time shifts. Shape is (number of sets, number of QSO images)
+print results.truetsarray # Idem, for the TRUE time shifts, in case of simulated data
+print results.qs # A 1D array with the "chi2" or dispersion values. Shape is (number of sets).
+
+# Note that these "tsarrays" contain time shifts, not time delays.
+# To get time delays between images "A" and "B" (i.e., results.labels[0] and results.labels[1]),
+# you will have to compute the differences yourself:
+
+measured_delays = results.tsarray[:,1] - results.tsarray[:,0]
+print measured_delays

pycs/regdiff2/gpr.py

@@ -6,7 +6,7 @@
 import numpy as np
 
 
-def regression(x, y, yerr, theta0, verbose=True):
+def regression(x, y, yerr, verbose=True, **kwargs):
 	"""
 	Give me data points
 	yerr is the 1sigma error of each y
@@ -23,8 +23,11 @@ def regression(x, y, yerr, theta0, verbose=True):
 #  		random_start=3, verbose=verbose)
 
 
-	gp = GaussianProcess(corr='squared_exponential',
-		theta0=theta0, random_start=1, nugget = 0.001, verbose=verbose)
+	mykwargs = {"theta0" : 1.0e2, "nugget" : 0.001}
+	mykwargs.update(kwargs)
+	#print mykwargs
+
+	gp = GaussianProcess(corr='squared_exponential', random_start=1, verbose=verbose, **mykwargs)
 
 
 	#gp = GaussianProcess(corr='squared_exponential',

pycs/regdiff2/rslc.py

@@ -74,7 +74,7 @@ def wtv(self, method = "weights"):
 
 
 
-def factory(l, theta0 = 1.0e5, pad=0.0, pd=2, plotcolour=None):
+def factory(l, pad=0.0, pd=2, plotcolour=None, **kwargs):
 	"""
 	Give me a lightcurve, I return a regularly sampled light curve, by performing some regression.
 	
@@ -115,7 +115,7 @@ def factory(l, theta0 = 1.0e5, pad=0.0, pd=2, plotcolour=None):
 	def meanprior(query):
 		return (0.0 * query + mean_mag)
 
-	regfct = gpr.regression(jds, mags, magerrs, theta0=theta0)
+	regfct = gpr.regression(jds, mags, magerrs, **kwargs)
 	(rsmags, rsmagerrs) = regfct(rsjds)
 
 	return rslc(rsjds, rsmags, rsmagerrs, pad, pd, timeshift=timeshift, name=name, plotcolour=plotcolour)

pycs/sim/draw.py

@@ -335,7 +335,7 @@ def draw(lcs, spline, shotnoise=None, shotnoisefrac=1.0, tweakml=None, scaletwea
 
 
 
-def multidraw(lcs, spline=None, optfctnots=None, onlycopy=False, n=20, npkl=5, simset="draw", simdir=None, shotnoise=None, shotnoisefrac=1.0, truetsr=0.0, tweakml=None, scaletweakresi=True, tweakspl=None, shuffle=True, verbose=True, trace=False):
+def multidraw(lcs, spline=None, optfctnots=None, onlycopy=False, n=20, npkl=5, simset="draw", simdir=None, shotnoise=None, shotnoisefrac=1.0, truetsr=8.0, tweakml=None, scaletweakresi=True, tweakspl=None, shuffle=True, verbose=True, trace=False):
 	"""
 	Even higher wrapper to produce mock + tweaked lightcurves, and save them into a directory (as pickle files),
 	in preparation for analysing them with :py:func:`pycs.sim.run.multirun`

sphinx/source/index.rst

@@ -14,6 +14,7 @@ To measure time delays with ``pycs``, you'll typically write a script calling so
 
 If you have already read our :doc:`papers<citing>`, you might want to proceed with :doc:`installation`, or the :doc:`tutorial/tutorial`. To get a quick first impression of how PyCS looks in practice, install PyCS and then go directly to the :doc:`tutorial/demo1`. These show you how to **reproduce** the figures from the method paper.
 
+.. warning:: Please also read :doc:`this<warning>`.
 
 Questions ?
 -----------
@@ -28,6 +29,7 @@ Contents
 .. toctree::
 	:maxdepth: 2
 
+	A word of warning <warning>
 	installation
 	tutorial/tutorial
 	citing

sphinx/source/warning.rst

@@ -0,0 +1,13 @@
+Warning & Disclaimer
+====================
+
+
+By *itself*, the use of PyCS does not guarantee realistic time-delay uncertainty estimates. All we have (blindly) demonstrated regarding the quality of PyCS uncertainty estimates assumes a careful generation of the simulated light curves. Keep in mind that the PyCS approach involves setting values or ranges for several parameters. In principle, the impact of these settings has to be assessed on each light curve set, as described in the PyCS paper (`Tewes et al. 2013 <http://dx.doi.org/10.1051/0004-6361/201220123>`_).
+
+.. warning:: In particular, obtaining uncertainty estimates by running on simulated light curves that all have a same true delay (i.e., setting ``truetsr = 0``, in PyCS terminology) can lead to **vastly underestimated error bars**.
+
+It is of **high importance** to use simulations with a range of plausible true time delays to tune and/or verify the accuracy and precision of time-delay estimators. Tests on simulations with only a single true time delay do not probe reliably the quality of a time-delay estimation, as many time-delay estimators are prone to responding unsteadily to the true delay. Do not hesitate to contact us (see front page of this documentation) in case of doubts!
+
+.. note:: Please try to avoid publishing time delays with overly optimistic uncertainty estimates. Seemingly accurate measurements might be propagated into unrealistic cosmological inferences, and harm the community. Be critical, and aim for the most comprehensive instead of the smallest error bar.
+
+Thanks!